WNN implementation in pure lua



Lua-wnn is a weightless n-tuple neural network implementation written in pure Lua. It's designed to be simple to use, both in terms of dependencies and usage, whilst still being performant enough for common use-cases. the library has been tested under lua 5.3 and 5.4 however development continues against lua 5.4 only (patches will be accepted for 5.3 support within reason).

The library supports multiple classes, with classes being able to be added dynamically to a classifier simply by training the classifier with data of that class. In addition it supports bleaching and can use an incremental bleaching algorithm (meaning that the bleaching threshold can be incremented automatically during classification).

To avoid the need to train the classifier each time it's used, classifiers can be serialised and deserialised. Currently this uses json as the storage format and thus the dkjson library is also required.

For installation and usage instructions please see the manual.